NASA Technical Reports Server (NTRS) 19950027324: Mineralogic variations in fluvial sediments contaminated by mine tailings as determined from AVIRIS data, Coeur D'Alene River Valley, Idaho

4 Pages·0.24 MB·English

by  NASA Technical Reports Server (NTRS)

N95-33745
MINERALOGIC VARIATIONS IN FLUVIAL SEDIMENTS
CONTAMINATED BY MINE TAILINGS AS DETERMINED FROM
AVIRIS DATA, COEUR D'ALENE RIVER VALLEY, IDAHO
W.H. Farrand 1and Joseph C. Harsanyi 2
1) Science Applications International Corporation
1710 Goodfidge Drive, P.O. Box 1303, M.S. E-7-3
McLean, VA 22102
2) Applied Signal and Image Technology Co.
9193 Rolling Meadow Run
Pasadena, MD 21122
1.0 INTRODUCTION
The success of imaging spectrometry in mineralogic mapping of natural terrains
(e.g., Clark et al., 1992) indicates that the technology can also be used to assess the
environmental impact of human activities in certain instances. Specifically, this paper
describes an investigation into the use of data from the Airborne Visible/Infrared Imaging
Spectrometer (AVIRIS) for mapping the spread of, and assessing changes in, the
mineralogic character of tailings from a major silver and base metal mining district.
The area under investigation is the Coeur d'Alene river valley in northern Idaho.
Mining has been going on in and around the towns of Kellogg and Wallace, Idaho since
the t880's. In the Kellogg-Smelterville Flats area, west of Kellogg, mine tailings were
piled alongside the South Fork of the Coeur d'Alene River. Until the construction of
railings ponds in 1968 much of these waste materials were washed directly into the South
Fork. The Kellogg-Smelterville area was declared an EPA Superfund site in 1983 and
remediation efforts are currently underway. Recent studies (e.g., Horowitz et al., 1992)
have demonstrated that sediments in the Coeur d'Alene River and in the northern part of
Lake Coeur d'Alene, into which the fiver flows, are highly enriched in Ag, Cu, Pb, Zn,
Cd, Hg, As, and Sb. These trace metals have become aggregated in iron oxide and
oxyhydroxide minerals and/or mineraloids. Reflectance spectra of iron-rich tailing
materials are shown in Figure 1. Also shown in Figure 1are spectra of hematite and
goethite. The broad bandwidth and long band center (near 1lain) of the Fe3÷crystal-field
band of the iron-rich sediment samples combined with the lack of features on the Fea÷-o 2
charge transfer absorption edge indicates that the ferric oxide and/or oxyhydroxide in these
sediments is poorly crystalline to amorphous in character. Similar features are seen in
poorly crystalline basaltic weathering products (e.g., palagonites).
The problem of mapping and analyzing the downriver occurences of iron rich
tailings in the Coeur d'Alene (CDA) River Valley using remotely sensed data is
complicated by the full vegetation cover present in the area. Because exposures of rock
and soil were sparse, the data processing techniques used in this study were sensitive to
detecting materials at subpixel scales. The methods used included spectral mixture
analysis (Adams et al., 1993) and a constrained energy minimization technique (Farrand
and Harsanyi, 1994a).
2.0 DATA
The CDA River Valley iscurrently under investigation by the Spokane field
office of the USGS (C. Smith, pers. comm.). In a cooperative endeavor between the
USGS, the EPA and the Naval Research Laboratory's HYDICE (Hyperspectral Digital
Imagery Collection Experimen0 Project Office an AVIRIS flight was scheduled over the
47
CDARiverValley.Datawascollected
on May 24, 1993 at 21:40 UTC. Incomplete
coverage of the target area led to a reflight on September 17 at 19:40 UTC. An earlier
paper (Farrand and Harsanyi, 1994b) presented apreliminary examination of two scenes
from the May collection. The current paper considers three scenes from the September
collection that cover an area extending down river from approximately Smelterville to
Killarney Lake (see Figure 2).
The reflectance of samples collected in the field were measured in the laboratory.
A subset of the samples were measured at Brown University's RELAB. The majority of
the samples were measured on the USGS Reston Beckman (UV5240).
The AVIRIS data were analyzed in aradiometrically calibrated format and also in
a format with DN equal to apparent surface reflectance. The conversion from radiance to
reflectance was done using both surface reflectance data and the ATREM (Gao et al.,
1993) radiative transfer based approach. First the data were converted to apparent surface
reflectance through the use of ATREM which effectively removed atmospheric variability.
This was followed by the application of a modified flat field (MFF) method (Farrand,
1992) applied to the ATREM corrected data which acted to remove instrumental noise that
is retained in the ATREM correction.
3.0 ANALYSIS
The AVIRIS radiance data were initially examined with spectral mature
analysis. An iterative approach utilizing several runs and successive examination of the
resulting RMS error images (Adams et al., 1993) suggested four endmembers: shade,
vegetation, an agricultural soil, and an iron oxide or oxyhydroxide rich soil. As was
discussed above, the iron-rich soil is the most likely candidate for bearing heavy metals
leached from the mine tailings. In the fraction image for this endmember, other materials
besides the ferruginous soil (for instance, the pavement on 1-90) appeared bright on the
fraction image.
In order to cut down on the false alarm rate, arecently developed detection
algorithm that has already shown good success in detecting distributed subpixel target
materials, the constrained energy minimization (CEM) technique (Farrand and Harsanyi,
1994b), was applied to the data. The key to the CEM technique is to determine a vector
operator, w composed of weights (wt...wm) that suppresses the unknown and undesired
background spectra while enhancing the target spectrum d. The summed pixel energy can
be represented by a scalar value Yi. The CEM operator is defined by two constraints.
The first constraint, for any given pixel, is to minimize the energy summed across the
wavelength range (e.g., minimize yi). The second constraint is that when applied to the
target spectrum, Yi= 1, e.g., wrd=l.
CEM was applied to subsections of the three AVIRIS scenes. The target
spectrum for input to the CEM routine was derived from the data itself. Three pixels
most like the BH-3 sample spectrum shown in Figure 1were identified. The radiance
spectra of these pixels were then extracted from a radiance data set that was subsectioned
to 188 bands to exclude several of the shortest and longest wavelength bands,
spectrometer overlap regions and atmospheric water absorptions at 1.4 and 1.9lain.
These radiance spectra were averaged and the average was used as input to the CEM
routine. As might be expected, the resulting CEM component images showed that the
iron-rich sediments are concentrated primarily along the CDA River. There were also
known concentrations of tailings in a dump west of Smelterville. Weaker responses are
returned from some fields in the CDA River Valley.
The CEM routine identified 159 pixels in the three scenes that were most like
the iron-rich sediments at Cataldo Flats. These pixels were then extracted from the
ATREM+MFF corrected data cube and transformed via a principal components analysis
(PCA). The PCA was applied in two parts. 88 VNIR channels were analyzed separately
48
from 45 SWIR channels. This was done because iron bearing minerals primarily have
features in the VNIR and clay minerals primarily have features in the SWIR. By
separating the analyses, the resulting intrinsic dimensionality of the data set could be kept
to a manageable number of endmembers (e.g., 3to 4). For purposes of brevity only the
VNIR analysis will be discussed here.
The fn:st three principal components of the VNIR analysis were examined
interactively using the "XGobi" data visualization program (Swayne et al., 1992). By
rotating the data cloud, spectrally extreme endmember pixels could be identified. Two
distinct linear trends revealed themselves outlining two ridges of a tetraheAron. Pixel
spectra from the ends of these trends (i.e., the vertices of the tetrahedron) are shown in
Figure 3. Although the PCA was done on only the first 88 channels, the full spectral
range is shown in Figure 3 since, as it turns out, there isextra information to be gained
from the SWIR region as well. Figure 3reveals two similar appearing, but nonetheless
distinct, sets of spectra. Both trends have one endmember with a strong "1 gin" Fe
crystal field absorption feature and one with only a weak absorption at that wavelength.
The spectra, t2eml and t2em2 appear to be wetter with stronger 1.4and 1.9 lain
absorption features. These spectra also display absorptions indicative of gypsum at 1.48,
1.53 and longwards of 1.7 lain. The presence of the sulfate mineral, gypsum, lends
further credence to the interpretation that the fermginous sediments are contaminated by
mine wastes.
4.0 CONCLUSIONS
These results provide encouragement that imaging spectrometer data can be used
effectively both to track the spread of fluvially distributed mine tailings and to provide
insight into changes in their mineralogic character. An enabling algorithm that aids
significantly in this process is the constrained energy minimization technique which, in
this instance, provided a unique identification of the ferruginous sediments associated with
mine waste contamination.
5.0 REFERENCES
Adams, J.B., M.O. Smith, and A.R. Gillespie,1993, Imaging spectroscopy:
Interpretation based on spectral mixture analysis, in Remote Geochemical Analysis:
Elemental and Mineralogical Composition, (C.M. Pieters and P.AJ. Englert, ed.s),
Cambridge University Press, New York, pp. 145-166.
Clark, R.N., G.A. Swayze, and A.J. Gallagher, 1992, Mapping the mineralogy and
lithology of Canyonlands, Utah with imaging spectrometer data and the multiple
spectral feature matching algorithm, inSummaries of the Third Annual JPL Airborne
Geoscience Workshop: AVIRIS JPL Publ. 92-14, 11-13.
Farrand, W.H., 1992, A comparison of methods for retrieving apparent surface reflectance
from hyperspectral dam. Proceedings of the International Symposium on Spectral
Sensing Research, pp.1154-1164.
Farrand, W.H. and J.C. Harsanyi, 1994a, Mapping distributed geological and botanical
targets through constrained energy minimization. Proc. Thematic Conf. Geol. Remote
Sens. lOth, 419-429.
Farrand, W.H. and J.C. Harsanyi, 1994b, An examination of mine waste contamination
in the Coeur D'Alene River Valley, Idaho using imaging spectrometer data. in Proc.
ISSSR '94 (in press).
Gao, B.-C., K.B. Heidebrecht, and A.F.H. Goetz, 1993, Derivation of scaled surface
reflectances from AVIRIS data. Rem. Sens. Env. 44, 165-178.
Horowitz, A., K.A. Elrick, and R.B. Cook, 1992, Effect of mining related activities on
the sediment trace element geochemistry of Lake Coeur d'Alene, Idaho, USA: Part 1.
Surface sediments, U.S. Geological Survey Open-file Report 92-109, 30 pp.
Swayne, D.F., D. Cook and A. Buja, 1992, User's manual for XGobi, Bellcore Technical
Memorandum
49
0.50 _ _--q--' -_- , , '_T'%'+ ' L ' _ _*t,- "_'_4
0.40 . +_
u 0.30
c
.°9
'_ 0.20
_ _ & goethile
,3 atoldo Fiats sample BH-3
0.10 ' , hemotite
£1
F
0.00 :
I I I _ _ _ , I i _ _ ,
0.5 10 1.5 2.0
Wavelength (in /.zm)
Figure 1. Laboratory reflectance spectra of iron-rich sediments collected near the Coeur d'Alene River
compared against library spectra of goethite and hematite.
_--. Coeur d'Alene
[ _t Coeur d'Alene River Valley
__i Orthern IdahO
Figure 2. Sketch map of Coeur d'Alene River Valley in northern Idaho. The AVIRIS flightlines axe
outlined bythe box.
500- i i i i
4OO
S" ._ , , .++
r'" _ ++ '_. \:;j
o
o
300 ,.,j,_ tleml/.t "j'" "/" "'_ = ._"._..._ •
E
c 2OO .i.:#./r _ - . ffs .V¢,. .,.t,.A'i :"_
g ..: j ' ; .' b.
I" i "IP-- ! ' rt
/'_tlem2 !7i t2eml _.
loo
/ l
O.b I.O 15 2 0
Wavelength (m|crometerq)
Hguz¢ 3. ]zon-rich sediment endmember_ deu=rmined fxom PeA of VNIR channels and
in_-active data analysis.
5O